TESTING OUT-OF-SAMPLE PERFORMANCE OF CORRELATION MODELS FOR PORTFOLIO CHOICE AND VALUE-AT-RISK

Our research strives to determine the relative out-of-sample performance of constant and dynamic correlation models in the context of portfolio choice and value-at-risk (VaR). We specify and estimate the dynamic conditional correlation (DCC) model proposed by Engle (2002) and the constant conditional correlation model proposed by Bollerslev (1990) and benchmark their performance against unconditional estimates. We use two data sets of daily returns comprised of three broad North American market indices and backtest the models over a period of more than 10 years which has not been done in the previous reviewed literature. Consistent with previous studies, we find DCC outperforms CCC and unconditional estimates, especially in times of changing volatility and correlation. VaR fails to adequately capture market risk during the 2008 financial crisis and the distribution assumption of innovations is more important than the choice of correlation model.